Cancer and Non-Cancer Controls in Studies on the Effect of Tobacco and Alcohol Consumption

©International Epidemiological Association 1991 Printed in Great Britain

Cancer and Non-Cancer Controls in

Studies on the Effect of Tobacco and

Alcohol Consumption

SALVATORE BARRA*, ANNA E BARON*,**, SILVIA FRANCESCHI*,t, RENATO TALAMINI* AND CARLO LA VECCHIAt,§

Barra S (Epidemiology Unit, Aviano Cancer Center, Via Pedemontana Occ, 33081 Aviano (PN), Italy), Bar6n A E, Franceschi S, Talamini R and La Vecchia C. Cancer and non-cancer controls in studies on the effect of tobacco and alcohol consumption. International Journal of Epidemiology 1991; 20: 845-851.

A comparison of risk estimates using controls with other cancers versus controls with acute diseases unrelated to tobacco and alcohol consumption in the study of the effect of these two factors has been performed using data on tumours of the oral cavity and pharynx from an ongoing case-control surveillance programme in Northeastern Italy. Similar results were obtained using either type of controls: as compared to never smokers, moderate smokers (=S14 cigarettes/day) showed age- and sex-adjusted odds ratio (OR) = 5.2 (95% confidence interval (Cl): 2.9-9.2) when using cancer controls and 5.8 (95% Cl: 3.3-10.1) when using non-cancer controls. Similarly, those who had smoked for 40 years or longer showed ORs of 7.4(95% Cl: 4.0-13.6) and 8.8(95% Cl: 4.9-15.6), respectively, using cancer and non-cancer controls. For moderate drinkers of alcoholic beverages (21-34 drinks/week) and heavy drinkers (2=84 drinks/ week) the ORs, as compared to individuals who drank <21 drinks/week, were 1.9 (95% Cl: 1.0-3.6) and 2.2 (95% Cl: 1.2-4.0) and 10.6(95% Cl: 5.5-20.6) and 11.4(95% Cl: 6.0-21.4) using cancer and non-cancer controls, respectively. The same comparability of ORs for tobacco- and alcohol-related variables using either type of controls was observed when separate analyses of the two sexes were performed. The close similarity between cancer and non-cancer controls in studies on tabacco- and alcohol-related risks may be exploited when the choice of other types of controls would increase the costs and the feasibility of the study, and thus hamper its statistical power. Moreover, this investigation provides some reassurance about the validity of risk estimates using carefully selected groups of hospital controls.

Tobacco and alcohol have been well established as the most important risk factors for tumours of the upper aero-digestive tract.u The influence of these habits has

been explored in several studies1"" using either

com-munity or hospital controls. In a few studies,12"15 a

cancer control group has been used and in one study16

results obtained with multiple control groups including cancer and non-cancer hospital controls have been shown.

The advantages of the use of cancer controls, which

•Epidemiology Unit, Aviano Cancer Center, Via Pedemontana Occ., 33081 Aviano (PN), Italy.

" D e p a r t m e n t of Preventive Medicine and Biometrics, University of Colorado Health Sciences Center, Denver, CO 80262, USA. fEpidemiology and Cancer Working Group, European Cancer Pre-vention Organization (ECP), Bruxelles, Belgium;

t'Mario Negri1 Institute for Pharmacological Research, Via Eritrea 62, 20157 Milan, Italy.

^Institute of Social and Preventive Medicine, University of Lausanne, 1005 Lausanne, Switzerland.

have been discussed by some authors,1718 include the

reduction of selection, recall and interviewer biases. However a potential disadvantage of their use is the inclusion of cancer types which are associated, posi-tively or negaposi-tively, with the exposures of interest, resulting in under- or overestimation of risk estimates. The purpose of the present investigation is to exam-ine the comparability of risk estimates for smoking and drinking habits in tumours of the oral cavity and phar-ynx using two different control groups, one selected from patients admitted to hospital for a wide spectrum of acute diseases unrelated to tobacco and alcohol use, and the other control group selected from patients with cancers also believed to be unrelated to these expo-sures. The results are expected to be useful in demon-strating that cancer controls constitute an appropriate control group in this context. More importantly, they will provide further information about the validity of risk estimates using hospital controls for two behav-ioural risk factors which have been shown to be less than reliably, and therefore inaccurately, reported in 845

INTERNATIONAL JOURNAL OF EPIDEMIOLOGY hospital interview settings.19 For this purpose we took

advantage of the data collected in the framework of a case-control surveillance programme ongoing since 1985 in Pordenone province, Northeastern Italy. MATERIALS AND METHODS

The study design of the present case-control surveil-lance has been described elsewhere.20 Briefly, 236 male

cases and 36 female cases below age 75 years (median age = 60), with histologically-confirmed cancer of the oral cavity and pharynx, diagnosed in the hospitals of the western part of the Friuli-Venezia Giulia region (Pordenone province) between January 1985 and July 1990, constituted the case group. All cases were resi-dent in the same area in which the hospitals are located.

Similarly, criteria for inclusion as a cancer control were: (1) to be below age 75 years (median age = 57) (2) to be affected by a cancer type not consistently related to tobacco or alcohol consumption and (3) to come from the same catchment area as the cases of cancer of the oral cavity and pharynx. After a review of recent epidemiological literature and our partly pub-lished case-control study results21"23 five tumour sites

were selected for inclusion as controls: colorectal, renal cell, prostate, thyroid, and haematological (i.e., Hodgkin's disease, non-Hodgkin's lymphoma and multiple myeloma). All eligible cancer controls were included, thus making a group in which each cancer site accounted for about one-fifth of the total. Table 1 shows the distribution of 577 males and 446 females according to these cancer sites.

Eligible non-cancer controls were individuals admit-ted for acute illnesses to selecadmit-ted departments of local hospitals. These departments were chosen in order to cover the same catchment area as those from where cancer patients originated and to allow the inclusion of a wide spectrum of diseases (i.e., no category was to account for more than a quarter of the total). Specific-ally excluded were patients who had malignant tumours or any condition consistently related to alcohol or tobacco consumption. A total of 1122 male controls and 762 female controls chosen on the basis of age (below 75 years, median age = 58), and area of residence were interviewed. Their distribution accord-ing to hospital discharge diagnostic categories is given in Table 1. In particular, non-traumatic orthopaedic conditions included mainly low back pain and disk dis-orders; traumatic conditions included mainly sprains and fractures; and surgical conditions were chiefly represented by benign cysts, abdominal hernia and acute appendicitis. The 'Others' category mostly in-cluded skin illnesses.

All study patients were interviewed in hospital. Refusals were about 3% for all three study groups. Interviewers were pilot-trained to reduce variability using the same pre-coded questionnaire to obtain information on sociodemographic factors, occupation, lifestyle, including tobacco and alcohol consumption habits, dietary habits, and past history of selected medical conditions. All of the information referred to the patient's situation before the onset of the disease which led to the hospital admission.

Information on smoking habits included smoking status (never, ex, current smoker), the type of products smoked (cigarettes, pipe or cigar), the number of ciga-rettes (or pipe/cigars) smoked per day before the onset of the disease, duration of cigarette smoking, the age at starting smoking, and, for ex-smokers, years since they quit smoking. In the part of the questionnaire dealing with alcohol consumption, the number of glasses of alcohol-containing beverages (wine, beer, spirits) con-sumed per week was elicited. Taking into account the different alcohol concentration, consumption of one drink corresponded to 150 ml of wine, 330 ml of beer and 30 ml of spirits (i.e. approximately 15 ml of ethanol).

The odds ratios (ORs), together with their 95% approximate confidence interval (CI) were computed accounting simultaneously for age (in quinquennia), sex, years of education, occupation and for the recipro-cal confounding effects of tobacco and alcohol. Uncon-ditional multiple logistic regression with maximum likelihood estimation was used to obtain these estimates.24

RESULTS

Table 2 shows the distribution of cases and two control groups according to sex and sociodemographic charac-teristics. When compared to either cancer or non-cancer controls, cases of non-cancer of the oral cavity and pharynx appeared to be less educated. Also the distri-bution by occupation was different and, in particular, farmers were more frequent among male cases than among either of the control groups (27% versus 18% of cancer controls and 14% of non-cancer controls), while fewer female cases than controls reported clerical/professional occupation (6% versus 20% in both control groups).

Table 3 gives the risks of cancer of the oral cavity and pharynx for various measures of tobacco consumption. A close similarity in the ORs using the two different types of controls was evident for all smoking-related variables. Moderate smokers (=£14 cigarettes/day) showed, as compared to never smokers, an OR of 5.2 (95% CI: 2.9-9.2) when contrasted to cancer controls

RISK ESTIMATES FOR TOBACCO AND ALCOHOL CONSUMPTION

TABLE I Distribution of cancer and non-cancer controls according to sex and diagnosis. Pordenone, Italy, 1985-1990

Males Colorectal cancer Kidney cancer Prostate cancer Haematological cancer Thyroid cancer Females Colorectal cancer Kidney cancer Haematological cancer Thyroid cancer Cancer controls (No. = 577) No. 130 82 125 195 45 Cancer controls (No. = 446) No. 82 51 173 140 (%) (22) (14) (22) (34) (8) (%) (18) (11) (39) (31) Orthopaedic Trauma Surgical conditions Eye disorders Others Orthopaedic Trauma Surgical conditions Eye disorders Others Non-cancer controls (No. = 1122) No. 251 291 303 67 210 Non-cancer controls (No. = 762) No. 243 151 131 53 183 (%) (22) (26) (27) (6) (19) (%) (32) (20) (17) (7) (24)

and 5.8 (95% CI: 3.3-10.1) when contrasted to non-cancer controls; the ORs associated with heavy smok-ing (=£25 cigarettes/day) were 9.6 (95% CI: 4.9-18.9) and 12.2 (95% CI: 6.4-23.2), respectively. Those who had smoked for 40 years or more had a 7.4-fold increased risk when compared to cancer controls and 8.8-fold increased risk when compared to non-cancer controls. Such similarities in the ORs were also obtained with age at starting smoking (OR = 6.8,95% CI: 3.8-12.2, and 6.6, 95% CI: 3.8-11.5, using cancer

and non-cancer controls, respectively, for =£16 versus 3^25 years) and, among ex-smokers, with years since quitting (OR = 3.9using both typesof controls for <10 years).

Drinking-related variables are shown in Table 4. For moderate wine drinkers (21-34 drinks/week) the ORs, as compared to individuals who drank <21 drinks/ week, were 1.8 (95% CI: 1.0-3.1) and 1.7 (95% CI: 1.0-3.1) using cancer and non-cancer controls respect-ively. Heavy wine drinkers (^84 drinks/week) had

TABLE 2 Distribution of cases of cancer of the oral cavity and pharynx, and cancer and non-cancer controls according to sex and various

socio-demographic characteristics* Pordenone, Italy, 1985-1990

Age (years) « 4 9 50-59 60-69 ^ 7 0 Education (years)** « 4 5-7 S=8 Marital status Never married Ever married Occupation** Clerical/Professional Manual worker Farmer Other Oral cavity-pharynx cancer cases Males No. 58 65 85 28 58 132 45 34 202 52 118 64 2 (%) (25) (28) (36) (12) (25) (56) (19) (14) (86) (22) (50) (27) (1) Females No. 7 15 11 3 15 16 5 2 34 2 10 4 20 (%) (19) (42) (31) (8) (42) (44) (14) (6) (94) (6) (27) (11) (56) Cancer controls Males No. 126 132 168 151 145 259 173 61 516 190 266 104 17 (%) (22) (23) (29) (26) (25) (45) (30) (11) (89) (33) (46) (18) (3) Females No. 164 110 95 77 142 178 126 62 384 90 108 74 174 (%) (37) (25) (21) (17) (32) (40) (28) (14) (86) (20) (24) (17) (39) Non-cancer controls Males No. 353 331 283 155 221 540 361 133 989 346 609 158 9 (%) (32) (29) (25) (14) (20) (48) (32) (12) (88) (31) (54) (14) 0) Females No. 313 177 152 120 241 333 187 102 660 149 221 109 282 (%) (41) (23) (20) (16) (32) (44) (24) (13) (87) (20) (29) (14) (37) 'Total sample size varies because of missing values.

TABLE3 Odds ratios for cancer of the oral cavity and pharynx according lo smoking habits.* Pordenone, Italy, 1985-1990 Never smoker Cigar or pipe smoker only Oral cavity cancer Males 6 5 Cigarette smoker (cigarettes/day)

€ 1 4 15-24 5=25 Xj trend 44 69 48

Duration of smoking (years) <30 30-39 5=40 X? trend 53 69 99

Age at starting smoking (years)

17-24 < 1 6 X* trend 24 113 91

Years since quitting smoking £=10 <10 X\ trend 22 41 pharynx cases Females 15 — 14 4 1 4 9 8 2 9 10 — 2 Cancer controls Males 115 7 86 98 40 161 99 171 73 245 141 139 94 Females 330 — 48 21 8 76 24 15 48 53 14 12 26 Odds r a t i o " (95% CI) 1 5.2 (1.4-19.6) 5.2 (2.9-9.2) 5.8 (3.2-10.5) 9.6 (4.9-18.9) 42.lt 2.7 (1.5-4.9) 7.0 (3.9-12.6) 7.4 (4.0-13.6) 55.3t 2.8 (1.4-5.4) 4.7 (2.7-8.1) 6.8 (3.8-12.2) 49.0f 1.4 (0.6-3.1) 3.9 (2.0-7.8) 18.8t Non-cancer controls Males 216 7 165 221 80 391 221 266 123 500 279 236 201 Females 553 — 89 47 , 7 146 37 22 69 83 52 25 38 Odds ratio* * (95% CI) 1 6.4 (1.8-23.6) 5.8 (3.3-10.1) 6.1 (3.5-10.9) 12.2 (6.4-23.2) 57.5t 2.7 (1.5-4.7) 6.9 (3.9-12.1) 8.8 (4.9-15.6) 75.9t 3.3 (1.7-6.2) 4.9 (2.9-8.4) 6.6 (3.8-11.5) 51.Of 1.6 (0.8-3.5) 3.9 (2.0-7.8) 19.2t 'Total sample size varies because of missing values.

"Estimates adjusted for age, sex, years of education, occupation and number of alcoholic drinks per week.

tP<0.01.

ORs of 13.7 (95% CI: 6.9-27.2) and 15.6 (95% CI: 8.2-29.7), respectively. Close similarities were observed for beer and spirits as well, although these alcoholic beverages were substantially less frequently used (Table 4). The same pattern of risk was obtained as regards the total alcohol intake (OR = 10.6, 95% CI: 5.5-20.6, and OR = 11.4, 95% CI: 6.0-21.4, using cancer and non-cancer controls, respectively).

Separate analysis of the two sexes (not shown) showed similar results, thus indicating that the close similarities in ORs based on either cancer or non-cancer controls apply to males or females.

DISCUSSION

Control selection is crucial in case-control studies,

since the use of inappropriate controls can lead to both selection and information bias and, therefore, affect the validity of a study.25 Cancer controls are commonly

utilized in studies performed in clinical settings, where establishing an internal comparison between patients with different, but generally closely related, cancer sites (e.g. cancers of the upper respiratory and diges-tive tract, gynaecological tumours, etc.) seen by the same physician saves both time and money.26 However,

the problem is that the similarities between such cases and controls, not only as regards the anatomical site but, in most instances, the probable aetiology, result in estimation of relative risks which are biased towards the null. Other limitations of using cancer controls, which must be borne in mind, include

representative-TABLE 4 Odds ratios for cancer of the oral cavity and pharynx according to drinking habits.* Pordenone, Italy, 1985-1990 849 Wine (drinks/week) 0-6 7-20 21-34 35-55 56-83 ^ 8 4 X; trend Beer (drinks/week) 0 1-13 5S14 X] trend Spirits (drinks/week) 0 1-13 X2, trend Total (drinks/week) 0-6 7-20 21-34 35-55 56-83 » 8 4 X] trend Oral cavity-pharynx cancer cases Males 5 7 26 42 95 61 141 27 68 107 66 63 3 4 17 29 78 105 Females 5 14 9 4 4 27 5 4 30 3 3 4 13 11 2 5 1 Cancer controls Males 41 84 176 119 120 37 325 144 108 275 217 85 30 62 127 136 133 89 Females 1461 217J 68 10 4 1 395 41 10 359 77 10 1401 200/ 75 19 10 2 Odds ratio** (95% CI) 1 1.8 (1.0-3.1) 3.5 (1.9-6.4) 7.6 (4.2-13.8) 13.7 (6.9-27.2) 89.6t 1 0.5 (0.3-0.9) 1.5 (1.0-2.2) 1.8 1 0.6 (0.4-0.8) 1.4 (1.0-2.2) 0.5 1 1.9 (1.0-3.6) 2.1 (1.1-4.2) 5.8 (3.0-11.1) 10.6 (5.5-20.6) 78.lt Non-cancer controls Males 82 138 329 272 240 61 633 242 247 554 391 177 62 107 215 291 260 187 Females 2671 359J 118 14 2 2 643 107 12 650 100 12 2541 347j 123 31 3 4 Odds ratio** (95% CI) 1 1.7 (1.0-3.1) 3.3 (1.8-5.9) 6.8 (3.9-12.1) 15.6 (8.2-29.7) 107.9t 1 0.7 (0.4-1.0) 1.4 (1.0-1.9) 1.5 1 0.8 (0.6-1.1) 1.6 (1.1-2.3) 1.1 1 2.2 (1.2-4.0) 2.4 (1.2-4.7) 6.6 (3.5-12.5) 11.4 (6.0-21.4) 92.8t •Total sample size varies because of missing values.

"Estimates adjusted for age, sex, years of education, occupation and smoking habits.

tP<0.01.

ness of risk estimates, additional potential confound-ing by exposure-related factors, and potentially different catchment areas across cancer types.18

A potential strength of the use of cancer controls, in case-control studies, is, however, the possibility of minimizing many problems related to other hospital and population controls, namely recall, selection and

information bias. These biases are often not quanti-fiable and may be towards or away from the null. Speci-fically, the choice of cancer controls has been advocated in the past to reduce recall bias, since all cancer patients should have been similarly interviewed for their past medical histories and various exposures and therefore should be more apt to recall them

accu-INTERNATIONAL JOURNAL OF EPIDEMIOLOGY rately, especially in the context of their severe

ill-ness.17'18 To this extent, cancer controls can be better

than non-cancer hospital controls and greatly superior to healthy community controls by providing risk esti-mates which are not biased away from the null value. In light of the practical difficulties in blinding the viewer to case-control status, a reduction in inter-viewer bias is also likely to be achieved by using cancer patients as a control group.17'18

Generally, in cancer epidemiology, hospital and population controls have been compared. The use of other cancer sites as controls has been evaluated less often. They were first utilized as far back as 1931.l2

Among case-control studies on risk factors for upper aero-digestive tract tumours, only a few studies'2'16

have used other cancer patients as controls. In one of them dealing with the importance of type of alcoholic beverages on oesophageal cancer risk, Tuyns et al.16

in-cluded digestive tract cancers and other medical and surgical patients in the control group. While the analyses performed by these authors did not separate the risk estimates by cancer versus non-cancer con-trols, the inclusion of cancer controls were presumably done to reduce recall and interviewer bias.27

Special concern arises when the aetiologies of differ-ent types of cancer (or other diseases), even if not closely linked anatomically, are unknown and, thus, similarities in risk factors cannot be confidently dis-carded.17'18 In this study we attempted to minimise this

problem by including control patients with cancers of different sites which, to the best of present epidemi-ological knowledge, are not consistently or substan-tially associated with tobacco and/or alcohol use. If they exist, some weak associations (e.g. direct for renal cancer28 and inverse for colorectal cancer29) may

coun-terbalance each other. Similarly, nearly a fifth of non-cancer controls came from each different category of disease, thus reducing the magnitude of selection bias. The effectiveness of this strategy is still open to debate,30"33 particularly in situations where

exposure-disease associations may be population-specific.33

However, tobacco and alcohol have been shown to be strong and consistent risk factors for oral cavity and pharynx cancer in a large variety of populations and settings.7~9:54'35 Their specificity would not then be

expected to bias the results in this study.

With regard to the specific exposures investigated in the present study, the work of Kelly et al.19 suggests

that the reliability of reporting alcohol consumption, and to a lesser extent tobacco use is only moderate in a hospital interview setting, thereby calling into ques-tion the validity of such self-reports. Differential mis-classification of specific risk factors across diagnostic

categories was not addressed in their study, but is ulti-mately of concern in the present study where both the cancer controls and the hospital controls are made up of several diagnostic groups. Given that the expected biases resulting from using cancer versus non-cancer controls are in opposite directions, the congruence between the risk estimates based on cancer controls and non-cancer controls in the present study suggests that the misclassification of alcohol and tobacco con-sumption was, in effect, non-differential.

Finally, when hospital records constitute the pre-dominant source of identification of cancer cases, a good comparability of catchment areas of cases with malignant diseases and controls with less severe acute conditions is very difficult to achieve. While this type of selection bias is unlikely to operate in the present study area, because of the established primary care and referral network, it is also likely to be alleviated by the choice of cancer controls.17'18

In this study the use of population controls would have been best in methodological terms, although the similarity of ORs when either cancer or non-cancer controls were used was so close as to provide, by itself, important information, particularly since all eligible and available cancer and non-cancer controls were included in the analysis. In practical terms, the present findings of high comparability of cancer and non-cancer controls in the study of tobacco- and alcohol-related tumours are useful given that, along with the minimization of selection, recall and information bias, the choice of cancer controls would diminish the costs and the difficulty of the study, and thus make it possible to increase its statistical power.

The use of other cancer types as controls may be also useful when investigating whether a specific exposure is associated with one specific tumour or to all tumours in general; or when a specific exposure, which has received particular attention in the media for its (adverse) effects, could result in overestimation because of recall bias. This could occur particularly in cancer studies on the effect of various components of diet which are subject to continuous attention in the media. Thus, while the present results strictly apply only to the effect of smoking and drinking on the risk of cancer of the oral cavity and pharynx, and are largely reassuring, they point to the need for further research on the validity of risk estimates obtaining using selected groups of hospital and non-hospital controls. ACKNOWLEDGEMENTS

The contribution of the Italian Association for Cancer Research, Milan and the Italian National Research Council (CNR Applied Projects 'Oncology', Contract

87.01544.44 and 'Risk Factors for Disease') are grate-fully acknowledged. We also thank Mrs Tiziana Ange-lin, and Dr Derna Gerdol for interviewing patients and Mrs Anna Redivo and Ilaria Calderan for editorial assistance.

REFERENCES

1IARC Working Group on the Evaluation of Carcinogenic Risks to Humans. IARC Monographs on the Evaluation of carcino-genic risks to humans. Tobacco smoking. Vol 38, Lyon, Inter-national Agency for Research on Cancer, 1986.

2 IARC Working Group on the Evaluation of Carcinogenic Risks to Humans. IARC Monographs on the Evaluation of carcino-genic risks to humans. Alcohol drinking. Vol. 44, Lyon, Inter-national Agency for Research on Cancer, 1988.

3 Wynder E L, Bross I J , Feldman R M. A study of etiological factors

in cancer of the mouth. Cancer 1957; 10: 1300-23. * Keller A Z, Terris M. The association of alcohol and tobacco with

cancer of the mouth and pharynx. Am J Public Health 1965; 55: 1578-85.

5 Doll R, Peto R. Mortality in relation to smoking: 20 years'

obser-vation on male British doctors. BrMedJ 1976; 2: 1526-36.

6 Mashberg A, Garfinkel L, Harris S. Alcohol as a primary risk factor

in oral squamous carcinoma. CA 1981; 31: 146-55.

7 Elwood J M, Pearson J C G, Skippen D H, Jackson S M. Alcohol,

smoking, social and occupational factors in the aetiology of cancer of the oral cavity, pharynx and larynx. Int J Cancer 1984; 34: 603-12.

* Blot W J, McLaughlin J K, Winn D M, elal. Smoking and drinking in relation to oral and pharyngeal cancer. Cancer Res 1988; 48: 3282-87.

' De Stefani E, Correa P, Oreggia F, et al. Black tobacco, wine and mat£ in oropharyngeal cancer. A case-control study from Uru-guay. Rev Epidem Santt Publ 1988; 36: 389-94.

10 Pgquignot G, Crosignani P, Terracini B, et al. A comparative study

of smoking, drinking and dietary habits in population samples in France, Italy, Spain and Switzerland. III. Consumption of alcohol. Rev Epidem Same Publ 1988; 36: 177-85.

11 Merletti F, Boffetta P, Ciccone G, Mashberg A, Terracini B. Role of

tobacco and alcoholic beverages in the etiology of cancer of the oral cavity/oropharynx in Torino, Italy. Cancer Res 1989; 49: 4919-24.

12 Hoffman F L. Cancer and smoking habits. Ann Surg 1931; 93:50-67.

" Schrek R, Baker L A, Ballard G P, Dolgoff S. Tobacco smoking as an etiologic factor in disease. I. Cancer. Cancer Res 1950; 10: 49-58.

"Feldman J G, Hazan M, Nagarajan M, Kissin B. A case-control investigation of alcohol, tobacco, and diet in head and neck cancer. Prev Med 1975; 4: 444-63.

15 Williams R R, Horm J W. Association of cancer sites with tobacco and alcohol consumption and socioeconomic status of patients: Interview study from the Third National Cancer Sur-vey. J N CI1977; 58: 525^47.

"Tuyns A J, Pequignot G, Abbatucci J S. Oesophageal cancer and

alcohol consumption: importance of type of beverage. Int J Cancer 1979; 23: 443-7.

17 Smith A H, Pearce N E, Callas P W. Cancer case-control studies with

other cancer as controls. Int J Epidemiol 1988; 17: 298-306.

18 Linet M S, Brookmeyer R. Use of cancer controls in case-control

cancer studies. Am J Epidemiol 1987; 125: 1-11.

" Kelly J P, Rosemberg L, Kaufman D W, Shapiro S. Reliability of personal interview data in a hospital-based case-control study. Am J Epidemiol 1990; 131: 79-90.

"Franceschi S, Talamini R, Barra S, et al. Smoking and drinking in relation to cancers of the oral cavity, pharynx, larynx, and eso-phagus in northern Italy. Cancer Res 1990; 50: 6502-7.

21 Franceschi S, Serraino D, Bidoli E, et.al. The epidemiology of non-Hodgkin's lymphoma in the north-east of Italy: a hospital-based case-control study. Leuk Res 1989; 13: 465-72.

22 Franceschi S, Fassina A, Talamini R, el al. Risk factors for thyroid

cancer in the northern part of Italy. Int J Epidemiol 1989; 18: 578-84.

23 Talamini R, Baron A E, Barra S, et al. A case-control study of risk

factors for renal cell cancer in northern Italy. Cancer Causes

and Control\990; 1: 125-31.

24 Mauritsen R. EGRET Software Program. Seattle, WA. Statistics and Epidemiology Research Corporation. 1986.

25 Schlesselman J J. Case-control Studies—Design, Conduct, Analysis. New York: Oxford University Press, 1982.

26 Brugere J, Guenel P, Leclerc A, Rodriguez J. Differential effects of tobacco and alcohol in cancer of the larynx, pharynx and mouth. Cancer 1986; 57: 391-5.

27 Jensen O M, Tuyns A J, P£quignot G. Usefulness of population

con-trols in retrospective studies of alcohol consumption. J Stud Alcohol 1978; 39: 175-82.

28 La Vecchia C, Negri E, D'Avanzo B, Franceschi S. Smoking and

renal cell carcinoma. Cancer Res 1990; 50: 5231-3.

29Sandier R S, Sandier D P, Comstock G W, Helsing D J, Shore D L.

Cigarette smoking and the risk of colorectal cancer in women. / JVC/1988; 80: 1329-33.

'"Lubin J H, Hartge P. Excluding controls: Misapplication in case-control studies. Am J Epidemiol 1984; 120: 791-3.

31 Pearce N, Checkoway H. Case-control studies using other diseases

as controls: problems of excluding exposure-related diseases. Am J Epidemiol 1988; 127: 851-56.

32 Wacholder S, Silverman D T. Re: 'Case-control studies using other

diseases as controls: problems of excluding exposure-related diseases'. Am J Epidemiol 1990; 132: 1017-8.

33 Pearce N, Checkoway H. The Author's Reply. Am J Epidemiol 1990; 132: 1018-9.

34 Brugere J, Guenel P, Leclerc A, Rodriguez J. Differential effects of

tobacco and alcohol in cancer of the larynx, pharynx, and mouth. Cancer 1986; 57: 391-5.

33 Franco E L, Kowalski L P, Oliveira B V, el al. Risk factors for oral

cancer in Brazil: a case-control study. Int J Cancer 1989; 43: 992-1000.